Assessing structural uncertainty in water–human systems: A multi-model ensemble approach for agricultural water management

Coupled hydro-economic models are increasingly used to support agricultural water management, yet their results can be highly sensitive to structural (model-form) choices in both human and water systems. We analyze how structural uncertainty propagates across a modular coupled human–water system model by building a hydrological multi-model ensemble for an irrigated catchment in Spain’s Douro basin (Tormes). The ensemble combines alternative hydrological models (Hydrologic Engineering Center’s Hydrologic Modeling System (HEC-HMS) and two Soil & Water Assessment Tool (SWAT) configurations) with four mathematical-programming representations of irrigator behavior (two Positive Mathematical Programming models and two Multi-attribute models) through a standardized coupling protocol that translates crop choices and irrigation demand into water-system inputs. We evaluate coupled performance of daily reservoir storage and harmonized diagnostics under two drought scenarios and four volumetric water charge levels based on standardized anomalies (z-scores). Results show that model structural uncertainty cascades through the ensemble and concentrates around irrigation onset, producing storage spreads of approximately 150–202 million m³ across coupled configurations, equivalent to 31–41% of Santa Teresa’s maximum permitted storage and 95–127% of annual downstream irrigation demand. Hydrological model choice explains most of this spread: family median differences reach 1.88–2.12 z-score units, meaning that the central tendency of standardized monthly storage anomalies shifts by almost two standard deviations depending on the hydrological structure selected. Economic-model structure mainly modulates within-family variability through heterogeneous water-demand signals. These findings imply that single-model decision support can underestimate the range of plausible futures and bias policy appraisal. We argue that systematically assessing model structural uncertainty through multi-model ensembles can strengthen robust decision-making on agricultural water management under drought and water charge interventions.